Do: understand the tools before being committed to a project
Don't: release with custom nonsense (As in stuff that would not make sense for portal or portal 2)
Hober wrote:
I'd say don't try and bit off more than you can chew. Work in small, iterative chunks. It's better to release a 10-puzzle mini-pack than make half a 40-puzzle pack and never finish or release it.
This is among the most important things you can learn regarding something like this.
To expand on that idea:
DON'T: Announce something big before you have a significant portion to show. Game developers know this, and you can learn from them. Most companies will not announce their next big project until it's well underway - in some cases, not until it's nearly complete. Setting out, you might be thinking "Yeah, I'm gonna do this! I have to tell everyone!" but you'd be wise to exercise restraint because chances are at some point you'll be thinking "This is just too much, I don't want to go on!" And if you made a big announcement right at the beginning, you'll find yourself with a bit of explainin' to do...
I speak from personal experience with this as well as from my own observations of others' failed projects.
Some of my own Do's and Don'ts:
DON'T: Be afraid to experiment. If you have an idea for something new, try it. If you can, produce a simple proof-of-concept for an idea for yourself in Hammer, then tweak it until you figure out how you might like to turn it into something more complete. If it doesn't work, no big deal - you learned something new, and it's time to move onto the next thing.
DON'T: Release everything you make in Hammer, or feel obligated to do so. Chances are, 90% of the .VMFs you produce won't go anywhere. And that's fine - they're used for trying new things, seeing if/how you can bring certain ideas to life, etc. Chances are if you release everything you make, it'll be all quantity and minimal quality. Experiment a lot, save your results, then apply them later to something release-worthy. It's like writing a paper for school - do a lot of research, keep all the chicken-scratch note pages for reference, then write a big final report summing it all up that gets turned in for the grade.
DON'T: Think that every map you produce has to be a mini-campaign. Sometimes a single puzzle is all you need! Remember that each of Valve's chambers is pretty short, yet they're each fun, both individually and as part of a larger whole. It's perfectly okay to have a map with only one challenge room instead of four or five. Simplicity is not your enemy. 
Finding a balance is still essential though - getting too simple leads to a dull puzzle just as getting too complicated leads to a frustrating one.
DO: Keep trying! Don't give up just because one map was a flop. It takes ages to get good at something, and map making is no exception. in fact, it's probably one of the hardest things out there to master. There are so many variables that need to be considered, and there's a lot to learn in the process, like which entities do what, etc...
DO: Playtest repeatedly! You might never even think of an exploit in your map that someone else immediately jumps to. You might also think your challenge is easy while someone else finds it mind-numbingly difficult or even impossible to figure out with what they're given. It is vital, especially in a puzzle game like Portal, to ensure that puzzles are doable and entertaining at the same time. Don't make it too obvious, or it's no fun. Don't make it too tough, or it's no fun. The balance is hard to locate sometimes, but it's there.
DO: Study and utilize what you like best about Valve's maps. Don't steal their stuff, but if you like what they've done somewhere, examine it closely to learn how they did it. Remember that everything you do is utilizing Valve-designed test elements. Also remember that the majority of people playing your map will be expecting a challenge similar to what they've already experienced, rather than something completely different from the Valve-made levels, and the introduction of too much "outside-the-box" thinking as part of a solution may intimidate and drive people away unnecessarily.
Do take example of elements you see ingame in the official portal campains.
Do take screenshots from the official maps, and study how certain things look. (like the space under a extended panel for instance, or how several styles of room are lit)
Do make use of indicator lines, show what buttons power what doors, show how the elements are connected, and keep it clean to avoid confusion.
Do study how to make instances, and use them! Why would you invent the wheel 30 times?
Ever had to create a dropper that dropped 3 different cubes depending on the input, save it as a instance, maybe you can use it later on! Be sure though that you dont recycle the same puzzle over and over again. Things to make into instances are moving panels, moving platforms (horizontal/vertical elevators), special droppers, vacuum tubes with animated props in them, chamber entry halls (especially handy to quicly start up a new map and begin drafting up a room not even a minute after starting hammer)
Do try to break your own rooms while playtesting them.
Try to lose or destroy props, try to jump out of the room, try to screw up automatic doors, try to run back and forth on random places in your map and see if and where you get stuck, try to finish the map without portals, try to finish the map without using certain puzzle elements, like funnels, try EVERYTHING, and if you can do something that gets you stuck... FIX it, if you discover a solution that requires more cunning or dexterity, and isn't directly the most "obvious" method of solving, leave it in.
Don't use too many puzzle elements at the same time, unless you are absolute sure of what you are doing.
Often, the more elements you bring into the puzzle, the easier it might get to "cheat" the puzzle in some simple way you havn't forseen. (like creating a map with sentries, funnels, bridges, panels, cubes, paint and more... all at the same time, in the same room/puzzle.)
You might have thought of a puzzle where players would have to juggle cubes around a fizzler field while standing on a switch while singing "Still Alive" backwards. But eventually players could just bypass the entire puzzle by using the excursion funnel you put in to cause some extra "confusion".
Don't think you are required to start and end every level with a elevator ride.
If you are planning on releasing a series of (lets say 5) maps, you could have one map end you up behind the scenes and have you load the next level in a closed off hallway (airlock style).
Don't hide solution critical elements of a puzzle until the puzzle is halfway solved.
Like if you would hide a excursion funnel (behind a panel), that would take you up to a high up exit. Players who havn't seen the funnel are going to try to figure out a way to get up to the out of reach exit, and attempting all kinds of impossibly complex flings, and get stuck.
Don't: Expect player's to be experts at playing Portal or Portal 2.
Do make the player feel good about him or herself after the test
Do put some effort into the map
Don't use brushes for decals
Don't: Put traps in your secret areas (Unless it's "really" worth it).
Do: Learn how to use panels properly. They should always be two units thick; that way, animations like the 90-degree ones are made predictable and on-grid.
Do: Get rid of that damn PotatOS! Of you're not sung her for the plot, she's unnecessary and annoying. It's as simp,e as creating a logic_playerproxy, then sending it an input OnMapSpawn with a delay of half a second of RemovePotatosFromPortalgun.
Do: Play other people's maps. Learn from their mistakes, steal their ideas, et cetera.
Don't: Plagiarise completely. When I say 'steal', I mean it in the loosest sense of the word. Your map should be original; if it is a rehash of someone else's chamber, it doesn't show off your skills.
Don't: Make use of glitches in your puzzles. It infuriates me when I play a map and the ONLY solution, as confirmed by the creator, is to use the floating laser cubes glitch or something similar. What happens when the glitch is fixed, eh? Your map is made redundant, never played again, and your time and effort has been wasted (although, every map you make does give you experience).
Rubrica wrote:
Do: Use the SilentDissolve input for slime! It is infuriating, by this point in time, to still hear that fizzle sound every time a cube falls into the water.
Even though Valve does not do this I have to agree with you 100% on that one. I does not make sense for a cube to fizzle in slime. It looks so much better if it just falls in.
Do: Playtest your own map before releasing. Some of the best ideas come to you when you are interacting with your own map.
Do: Get others to playtest your map. They will give you a fresh perspective on the puzzles that never occurred to you.
DON'T: Put unavoidable turret ambushes into your map. By this I mean places where the player is required (or all-but-required) to die and respawn to even realize that turrets are present in the map. In general, if it's not possible under reasonable circumstances (i.e. without prior knowledge of the puzzle's layout or split-second precision reflexes) to complete a map without taking bullet damage from a turret, it's likely you need to redesign that portion of your level.
DON'T: Place turrets solely for the sake of placing turrets. If a puzzle is complete without turrets, it does not need them. Putting an ambush that must be diffused before the "real" puzzle can even be begun is redundant and un-fun for the player. Chances are someone is playing your map to solve a cognitive puzzle, not to practice split-second reaction to bullets. If someone wants that, they should go play a genuine FPS, not a first-person puzzler.
DON'T: Place any turrets that cannot be killed or avoided in any way. Invincible turrets, while a tough feat to pull off successfully, can be made to work well if you ensure that there is a way to avoid them without taking damage.
DON'T: Use Super Damage on turrets. EVER.
iWork925 wrote:
Dont: Make a laser relay(s) your only puzzle. Relay puzzles are boring.
The only good laser puzzles are ones that include other test elements, like the death field/moving platform chamber late in the game.
You'll never find me making a map with just straightforward laser puzzles. Most of my maps include lasers, but they are used in innovative ways.
P.S. In fact, too innovative by some people's standards.